Visual saliency-based landslide identification using super-resolution remote sensing data

Abstract

Landslides, ubiquitous geological hazards on steep slopes, present formidable challenges in tropical regions with dense rainforest vegetation, impeding accurate mapping and risk assessment. To address this, we propose an innovative deep-learning framework utilizing visual saliency for automatic landslide identification, employing super-resolution remote sensing image datasets. Unlike conventional models relying on raw images, our method leverages saliency-generated feature maps, achieving a remarkable 94% accuracy, surpassing existing models by 5%. Comprehensive experimental findings consistently demonstrate its superiority over established algorithms, highlighting its robust performance. This novel approach introduces a valuable dimension to landslide detection, particularly in complex terrains, offering a promising tool for advancing risk assessment and management in landslide-prone areas.